๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

AI/๋…ผ๋ฌธ ๋ฆฌ๋ทฐ1

VGGNet (ICLR 2015) ์š”์•ฝ ๋ฐ ๋ฆฌ๋ทฐ Very Deep Convolutional Networks for Large-Scale Image Recognitionhttps://arxiv.org/abs/1409.1556   I. Introductionํ•ด๋‹น ๋…ผ๋ฌธ์€ ๋ชจ๋ธ์˜ ๋ ˆ์ด์–ด๋ฅผ ๋” ๊นŠ๊ฒŒ ๋งŒ๋“œ๋Š” ๊ฒƒ์— ์ค‘์ ์„ ๋‘์—ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ๋‹ค๋ฅธ ํŒŒ๋ผ๋ฏธํ„ฐ๋“ค์„ ๊ณ ์ •ํ•˜๊ณ  3x3 ํฌ๊ธฐ์˜ ์ž‘์€ ํ•„ํ„ฐ๋กœ ๋ ˆ์ด์–ด๋ฅผ ์Œ“์•„ ๋ชจ๋ธ์˜ ๊นŠ์ด๋ฅผ ํš๊ธฐ์ ์œผ๋กœ ๋Š˜๋ ธ๋‹ค=> ๊ทธ ๊ฒฐ๊ณผ ๋‹น์‹œ์˜ SOTA ๋‹ฌ์„ฑ ๋ฐ ๋‹ค์–‘ํ•œ ๋ฐ์ดํ„ฐ์…‹์— ์ ์šฉ ๊ฐ€๋Šฅํ•œ ์„ฑ๋Šฅ ์ข‹์€ ๋ชจ๋ธ๋กœ ์†๊ผฝํžˆ๊ฒŒ ๋˜์—ˆ๋‹ค II. ConvNet Configurations2.1 Architectureํ›ˆ๋ จ์…‹์˜ ๊ฐ ํ”ฝ์…€์—์„œ ํ‰๊ท  RGB๋ฅผ ๋บ€ ์ „์ฒ˜๋ฆฌ๋ฅผ ํ•œ 224x224x3  RGB ์ด๋ฏธ์ง€ ์‚ฌ์šฉ์ „์ฒ˜๋ฆฌ๋œ ์ด๋ฏธ์ง€๋“ค์€ 3x3 ํฌ๊ธฐ์˜ ํ•„ํ„ฐ๊ฐ€ ์žˆ๋Š” ํ•ฉ์„ฑ๊ณฑ์ธต์„ ์ง€๋‚œ๋‹ค.. 2024. 1. 29.